CN109961435A - Brain image acquisition methods, device, equipment and storage medium - Google Patents
Brain image acquisition methods, device, equipment and storage medium Download PDFInfo
- Publication number
- CN109961435A CN109961435A CN201910260795.2A CN201910260795A CN109961435A CN 109961435 A CN109961435 A CN 109961435A CN 201910260795 A CN201910260795 A CN 201910260795A CN 109961435 A CN109961435 A CN 109961435A
- Authority
- CN
- China
- Prior art keywords
- brain image
- sample
- network model
- neural network
- obtains
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10104—Positron emission tomography [PET]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10108—Single photon emission computed tomography [SPECT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30016—Brain
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Nuclear Medicine (AREA)
Abstract
The embodiment of the invention discloses a kind of brain image acquisition methods, device, equipment and storage mediums.This method comprises: obtaining the first brain image of subject based on PET-Positron emission computed tomography technology or Single photon emission computer tomography technology;It obtains and complete brain image has been trained to obtain neural network model;First brain image is input to brain image and obtains neural network model, determines the corresponding deformation field of the first brain image;The first brain image is converted according to deformation field, the second brain image is obtained, using the second brain image as the brain image of subject.The technical solution of the embodiment of the present invention solves the problems, such as the limitation of process of image registration bring precision, improves the structure precision of the brain image of subject;Moreover, subject is not necessarily to carry out additional MR scanning, it is more convenient to operate.
Description
Technical field
The present embodiments relate to Medical Image Processing more particularly to a kind of brain image acquisition methods, device, equipment
And storage medium.
Background technique
Based on PET-Positron emission computed tomography imaging (Positron Emission Computed Tomography,
PET) technology, or, Single photon emission computer tomography (Single-Photon Emission Computed
Tomography, SPECT) technology, the brain image of acquisition can mention in terms of the early diagnosis of the diseases such as alzheimer, Kieren Perkins
For important information.Particularly, under the premise of the physiological structure of known brain image, different brains can be directly obtained from image
The ingestion of medicines value of position, with the development trend of predictive disease.In order to realize the above functions, a kind of common scheme is to subject
Brain image converted, to realize the construction standard of brain image, i.e., the location criteria of target pixel points in realization brain image
Change.
Using the brain image of above-mentioned acquisition as the first brain image, above-mentioned transformed brain image is as the second brain image.By
Structural information in the first brain image is less, and picture noise is higher, and resolution ratio is poor, and tracer is many kinds of, because
This is difficult the second brain image that structure more standard is directly acquired according to the first brain image.To solve the above-mentioned problems, existing side
Case in terms of Brian Imaging usually can wrap magnetic resonance imaging (the Magnetic Resonance containing more structural information
Imaging, MR) image is medium, realize the effect that the second brain image is obtained according to the first brain image and MR image.
Specifically, obtaining the first brain image and MR image of the brain of the same subject;According to traditional image registration
Method or manual registration method determine the deformation field (deformation field) of MR image, that is, determine MR image and second
Deformation field between brain image;First brain image is registrated with MR image, and above-mentioned deformation field is applied to after being registrated
First brain image obtains the second brain image.
The MR image but above-mentioned technical proposal places one's entire reliance upon, it is likely that need subject to receive additional MR scanning, very
Difficulty is applied to more pervasive and more cheap PET-CT system.Moreover, the registration of existing first brain image and MR image
Journey is easy to appear problem, directly affects the structure precision of the second brain image.
Summary of the invention
The embodiment of the invention provides a kind of brain image acquisition methods, device, equipment and storage mediums, additional not increasing
MR image on the basis of, improve the structure precision of the brain image of the subject got.
In a first aspect, may include: the embodiment of the invention provides a kind of brain image acquisition methods
Subject is obtained based on PET-Positron emission computed tomography technology or Single photon emission computer tomography technology
The first brain image;
It obtains and complete brain image has been trained to obtain neural network model;
First brain image is input to brain image and obtains neural network model, determines the corresponding deformation field of the first brain image;
The first brain image is converted according to deformation field, the second brain image is obtained, using the second brain image as subject
Brain image.
Optionally, above-mentioned brain image acquisition methods can also include:
The sample brain image for obtaining sample object, sample brain image and preset standard brain image are registrated, obtained
Sample deformation field, and using sample brain image and sample deformation field as one group of training sample;
Original neural network model is trained based on multiple training samples, brain image is obtained and obtains neural network mould
Type, wherein original neural network model includes convolutional neural networks model.
Optionally, above-mentioned brain image acquisition methods can also include:
Magnetic resonance samples brain image and PET-Positron emission computed tomography sample brain image based on sample object;And
Preset standard brain image, is trained original neural network model, obtains brain image and obtains neural network model;Or,
Magnetic resonance samples brain image and Single photon emission computer tomography sample brain image based on sample object;And
Preset standard brain image, is trained original neural network model, obtains brain image and obtains neural network model.
Optionally, training sample may include the first training sample and the second training sample;Correspondingly, described based on multiple
Training sample is trained original neural network model, obtains brain image and obtains neural network model, may include:
Original neural network model is trained based on multiple second training samples, preliminary brain image is obtained and obtains nerve
Network model;
Neural network model is obtained to preliminary brain image based on multiple first training samples to be trained, and is obtained brain image and is obtained
Take neural network model.
Optionally, sample object may include first sample object;First training sample can be as follows
It obtains:
The first sample is obtained based on PET-Positron emission computed tomography technology or Single photon emission computer tomography technology
The first sample brain image of this object, and, the magnetic resonance samples brain of first sample object is obtained based on mr imaging technique
Image;
Magnetic resonance samples brain image and preset standard brain image are registrated, obtain first sample deformation field, and will
First sample brain image and first sample deformation field are as one group of first training sample.
Optionally, sample object may include the second sample object;Second training sample can obtain as follows:
The second sample is obtained based on PET-Positron emission computed tomography technology or Single photon emission computer tomography technology
The second sample brain image of this object;
Second sample brain image and preset standard brain image are registrated, obtain the second sample deformation field, and by
Two sample brain images and the second sample deformation field are as one group of second training sample.
Optionally, using sample brain image and sample deformation field as one group of training sample after, above-mentioned brain image obtains
Method can also include:
The training sample set being made of multiple training samples is expanded based on preset data set amplification algorithm;
Correspondingly, described be trained original neural network model based on multiple training samples, brain image acquisition is obtained
Neural network model, comprising:
Original neural network model is trained based on the training sample set expanded, brain image is obtained and obtains nerve net
Network model.
Optionally, the training sample set being made of multiple training samples is expanded based on preset data set amplification algorithm
Increase, may include:
According at least two sample deformation fields that training sample is concentrated, and, preset random deformation field generates amplification and becomes
Shape field;
According to every group of training sample that training sample is concentrated, acquisition has converted sample brain image, and according at least two
Transformation sample brain image generation has converted amplification brain image;
According to amplification brain image and amplification deformation field has been converted, non-transformed amplification brain image is generated.
Optionally, above-mentioned brain image acquisition methods can also include:
Deformation field is adjusted according to the adjustment operation of the user's input received, and based on deformation field adjusted and
First brain image obtains neural network model to brain image and is trained again, and updates brain image according to training result and obtain mind
Through network model.
Optionally, the first brain image is converted according to deformation field, obtains the second brain image, may include:
It is converted according to position coordinates of the deformation field to the target pixel points in the first brain image, obtains the second mind map
Picture.
Second aspect, the embodiment of the invention also provides a kind of brain image acquisition device, the apparatus may include:
First brain image obtains module, by based on PET-Positron emission computed tomography technology or single photon emission
The first brain image of calculation machine tomographic techniques acquisition subject;
Model obtains module, has trained complete brain image to obtain neural network model for obtaining;
Deformation field determining module obtains neural network model for the first brain image to be input to brain image, determines first
The corresponding deformation field of brain image;
Brain image obtains module, for converting according to deformation field to the first brain image, obtains the second brain image, by the
Brain image of two brain images as subject.
The third aspect, the embodiment of the invention also provides a kind of equipment, which may include:
One or more processors;
Memory, for storing one or more programs;
When one or more programs are executed by one or more processors, so that one or more processors realize following behaviour
Make:
Subject is obtained based on PET-Positron emission computed tomography technology or Single photon emission computer tomography technology
The first brain image;
It obtains and complete brain image has been trained to obtain neural network model;
First brain image is input to brain image and obtains neural network model, determines the corresponding deformation field of the first brain image;
The first brain image is converted according to deformation field, the second brain image is obtained, using the second brain image as subject
Brain image.
Fourth aspect, the embodiment of the invention also provides a kind of computer readable storage mediums, are stored thereon with computer
Program realizes following operation when the computer program is executed by processor:
Subject is obtained based on PET-Positron emission computed tomography technology or Single photon emission computer tomography technology
The first brain image;
It obtains and complete brain image has been trained to obtain neural network model;
First brain image is input to brain image and obtains neural network model, determines the corresponding deformation field of the first brain image;
The first brain image is converted according to deformation field, the second brain image is obtained, using the second brain image as subject
Brain image.
The technical solution of the embodiment of the present invention, by the way that the first brain image, that is, PET image or SPECT image are input to brain
Image obtains in neural network model, determines the corresponding deformation field of the first brain image;The first brain image is carried out according to deformation field
Transformation obtains the second brain image, using the second brain image as the brain image of subject.Above-mentioned technical proposal solves image registration
The problem of process bring precision limits, improves the structure precision of the brain image of subject;Moreover, subject is without carrying out volume
Outer MR scanning, it is more convenient to operate.
Detailed description of the invention
Fig. 1 is the flow chart of one of embodiment of the present invention one brain image acquisition methods;
Fig. 2 is the flow chart of one of embodiment of the present invention two brain image acquisition methods;
Fig. 3 is that the structure of the original neural network model in one of embodiment of the present invention two brain image acquisition methods is shown
It is intended to;
Fig. 4 is the flow chart of one of embodiment of the present invention three brain image acquisition methods;
Fig. 5 is the structural block diagram of one of embodiment of the present invention four brain image acquisition device;
Fig. 6 is the structural schematic diagram of one of the embodiment of the present invention five equipment.
Specific embodiment
Invention is further described in detail with reference to the accompanying drawings and examples.It is understood that described herein
Specific embodiment be used only for explaining the present invention rather than limiting the invention.It also should be noted that for the ease of
It describes, only the parts related to the present invention are shown rather than entire infrastructure in attached drawing.
Before introducing the embodiment of the present invention, first the application scenarios of the embodiment of the present invention are illustrated.Neural network mould
Type is to carry out abstract computation model to the neuroid of human brain from the angle of information processing, it is therefore intended that allows computer can
The form of thinking for simulating human brain is some abstract to solve the problems, such as.Neural network model is a kind of end to end model, i.e. computer
It can learn useful feature automatically, to reduce complexity and incompleteness caused by artificial extraction feature.
But in the process field of field of image processing, especially PET image and SPECT image, due to neural network mould
The black box characteristic of type makes the bad judgement of correctness of the image obtained, and the image of the mistake got is possible to miss
It the malpractices such as examines, limits neural network model in the application of PET image and the process field of SPECT image.However, this hair
The application scenarios of bright embodiment can overcome the above problem completely, because, if being deposited by the deformation field that neural network model obtains
In error, error is certainly existed based on the transformed brain image of deformation field, this error is can to sentence by observing Direct Recognition
Disconnected.
Embodiment one
Fig. 1 is a kind of flow chart of the brain image acquisition methods provided in the embodiment of the present invention one.The present embodiment is applicable
In obtain subject construction standard brain image the case where.This method can be obtained by brain image provided in an embodiment of the present invention
Device is taken to execute, which can be realized that the device can integrate in various user terminals by the mode of software and/or hardware
Or on server.
Referring to Fig. 1, the method for the embodiment of the present invention specifically comprises the following steps:
S110, based on PET-Positron emission computed tomography technology or Single photon emission computer tomography technology obtain by
The first brain image of inspection person.
Wherein, the first brain image can be subject and carry out the image obtained in real time after head scanning, be also possible to from brain
The image obtained in image data base.First brain image can be PET image, can be SPECT image.Due to PET image and
SPECT image is functional diagram picture, and the physiologic information in image is more, and structural information is less.Then, according to PET image or
SPECT image is more difficult those of to directly determine in image which part that physiologic information belongs in brain structure is tied by traditional algorithm
Structure.
S120, acquisition have trained complete brain image to obtain neural network model, and the first brain image is input to brain image
Neural network model is obtained, determines the corresponding deformation field of the first brain image.
Wherein, the input that brain image obtains neural network model is the first brain image, i.e. PET image or SPECT image,
Output is the corresponding deformation field of the first brain image.Then, the first brain image obtained after subject carries out head scanning can be straight
It connects and is input to brain image neural network model, automatically determine deformation field corresponding with the first brain image.Brain image obtains nerve net
The application of network model allows and directly determines corresponding deformation field according to PET image or SPECT image, is not necessarily to subject
Receive additional MR scanning;It avoids because of PET image and MR image, or, the mistake of SPECT image and MR image in registration process
The lower situation of the structure precision of transformed brain image caused by accidentally.
S130, the first brain image is converted according to deformation field, obtain the second brain image, using the second brain image as by
The brain image of inspection person.
Wherein, the first brain image can be converted according to deformation field, obtains the second brain image, the second brain image is
Transformed first brain image.Since the second brain image is the standardized image of brain structure, each object pixel in image
Point is all located in its corresponding brain structure;Moreover, the pixel value of target pixel points can reflect physiologic information, then according to second
Brain image those of can directly determine in image which part-structure that physiologic information belongs in brain.
Optionally, it after obtaining deformation field D and the first brain image X, can be obtained according to image transformation functions Y=f (X, D)
Second brain image Y.The process of above-mentioned image transformation may be considered the conversion process of the position coordinates of target pixel points, i.e. basis
Deformation field converts the position coordinates of the target pixel points in the first brain image, obtains the second brain image.Illustratively, when
When the target pixel points of first brain image a-quadrant transform to the second brain image B area, only the position coordinates of target pixel points are sent out
It changes and changes, and the numerical value of target pixel points does not convert.Above-mentioned steps setting is advantageous in that, can automatically determine first
The corresponding target area of each target pixel points, obtains standardized second brain image of brain structure in brain image.
The technical solution of the embodiment of the present invention, by the way that the first brain image, that is, PET image or SPECT image are input to brain
Image obtains in neural network model, determines the corresponding deformation field of the first brain image;The first brain image is carried out according to deformation field
Transformation obtains the second brain image, using the second brain image as the brain image of subject.Above-mentioned technical proposal solves image registration
The problem of process bring precision limits, improves the structure precision of the brain image of subject;Moreover, subject is without carrying out volume
Outer MR scanning, it is more convenient to operate.
A kind of optional technical solution, above-mentioned brain image acquisition methods specifically can also include: the magnetic based on sample object
The sample brain image that resonates and PET-Positron emission computed tomography sample brain image;And preset standard brain image, to original
Neural network model is trained, and is obtained brain image and is obtained neural network model;Or the magnetic resonance samples based on sample object
Brain image and Single photon emission computer tomography sample brain image;And preset standard brain image, to original neural network
Model is trained, and is obtained brain image and is obtained neural network model.
Wherein, it is that training obtains on the basis of original neural network model that brain image, which obtains neural network model,.By
Functional imaging technology in PET technology and SPECT technology, obtained PET-Positron emission computed tomography sample brain image or
Be Single photon emission computer tomography sample brain image structural information it is less.Believe correspondingly, can wrap containing more structure
The magnetic resonance samples brain image of breath has great advantages during image converts, however, this needs subject to receive volume
Outer MR scanning, or use PET/MR instrument costly.
To solve the above-mentioned problems, original neural network model can be trained based on magnetic resonance samples brain image,
In other words, be original neural network model is trained based on the corresponding sample deformation field of magnetic resonance samples brain image so that
It has trained complete brain image to obtain in neural network model and has had structural information.That is, by PET-Positron emission computed tomography
As input, magnetic resonance samples brain image is corresponding for sample brain image or Single photon emission computer tomography sample brain image
Deformation field is trained original neural network model as output.Above-mentioned steps setting is advantageous in that, according to having trained
Standby brain image, which obtains in the deformation field that neural network model obtains, has structural information.In practical applications, even if the first brain
Image is the less PET image of structural information or SPECT image, is used as auxiliary due to having structural information in deformation field, can
To improve the structure precision of the second brain image got.
A kind of optional technical solution, above-mentioned brain image acquisition methods specifically can also include: according to the user received
The adjustment operation of input is adjusted deformation field, and obtains mind to brain image based on deformation field adjusted and the first brain image
It is trained again through network model, and brain image is updated according to training result and obtains neural network model.
Wherein, in the training process of original neural network model, training sample possibly can not cover the mind map of all situations
Picture, so that training complete brain image to obtain neural network model may need to be trained again.Then, mind is obtained in brain image
Application process through network model, when undesirable situation occurs in the second brain image of acquisition, equipment can provide operation circle
Face, so that user is adjusted the deformation field of acquisition or the second brain image.Illustratively, when user thinks the second brain image
In hippocampus position it is wrong when, can control mouse and adjust hippocampus to correct position.
Correspondingly, can be operated according to the adjustment to deformation field when equipment receives the adjustment operation of user's input
It is adjusted, for example, equipment can will deform when the a-quadrant in the second brain image is adjusted 5 pixels to the left by user
The region A1 corresponding with a-quadrant adjusts 5 pixels to the left in, with the deformation field after being adjusted.It in turn, can be with base
Neural network model is obtained to brain image in deformation field adjusted and the first brain image to be trained again, and is tied according to training
Fruit is updated.
By foregoing teachings it is found that when the first brain image in the prior art and MR image registration process when the error occurs,
Even if user is adjusted the second brain image, the first later similar brain image still will appear mistake.However, the present invention is real
The technical solution for applying example, when inputting the first similar brain image again, updated brain image has obtained neural network model
The first brain image that can handle this type, directly exports ideal deformation field, enhances the resolution to similar image.
Embodiment two
Fig. 2 is a kind of flow chart of the brain image acquisition methods provided in the embodiment of the present invention two.The present embodiment is with above-mentioned
It is optimized based on each technical solution.In the present embodiment, optionally, above-mentioned brain image acquisition methods, can also specifically wrap
It includes: obtaining the sample brain image of sample object, sample brain image and preset standard brain image are registrated, obtain sample change
Shape field, and using sample brain image and sample deformation field as one group of training sample;Based on multiple training samples to original nerve net
Network model is trained, and is obtained brain image and is obtained neural network model, the original neural network model may include convolution mind
Through network model.Wherein, details are not described herein for the explanation of term identical or corresponding with the various embodiments described above.
As shown in Fig. 2, the method for the present embodiment can specifically include following steps:
S210, the sample brain image for obtaining sample object, sample brain image and preset standard brain image are registrated,
Sample deformation field is obtained, and using sample brain image and sample deformation field as one group of training sample.
Wherein, sample brain image can be PET image, can be SPECT image, can also be MR image;Preset mark
Quasi- brain image can be MR image.And for different sample objects, preset standard brain image can be identical mind map
Picture is also possible to different brain images.For example, can be selected according to personal attribute informations such as the age of sample object, nationality, genders
Select corresponding standard brain image.
The registration Algorithm of sample brain image and standard brain image can be gradient descent method, optical flow method, block registration method etc..Sample
After the completion of this brain image and standard brain image registration, available sample deformation corresponding with sample brain image field, every group of sample
Brain image and sample deformation field can be used as one group of training sample.It repeats the above steps, available multiple groups training sample.
S220, original neural network model is trained based on multiple training samples, obtains brain image and obtains nerve net
Network model, wherein original neural network model includes convolutional neural networks model.
Wherein, using the sample brain image in training sample as input, sample deformation field is as output, to original nerve net
Network model is trained, and is obtained brain image and is obtained neural network model.Specifically, when original neural network model includes convolution mind
When through network model, illustratively, as shown in figure 3, may include multiple convolutional layers, pond in the centre of input layer and output layer
Change layer (being not drawn into figure) and full articulamentum.Each convolution operation can correspond to the filter of different number, particularly, by
What it is in output layer output is the deformation field for having 3 direction variable values, the last one convolution operation can be equivalent to 3 filtering
Device.It is to be appreciated that the element in each convolutional layer and full articulamentum in convolutional neural networks model is known variables, on
The process for stating model training is exactly to estimate the process of the optimal solution of these known variables.
S230, based on PET-Positron emission computed tomography technology or Single photon emission computer tomography technology obtain by
The first brain image of inspection person.
S240, acquisition have trained complete brain image to obtain neural network model, and the first brain image is input to brain image
Neural network model is obtained, determines the corresponding deformation field of the first brain image.
S250, the first brain image is converted according to deformation field, obtain the second brain image, using the second brain image as by
The brain image of inspection person.
The technical solution of the embodiment of the present invention is based on using sample brain image and sample deformation field as one group of training sample
Multiple groups training sample is trained original neural network model, and obtained brain image, which obtains neural network model, can cover more
More situations, completeness are more preferable.
A kind of optional technical solution, training sample may include the first training sample and the second training sample;Correspondingly,
Original neural network model is trained based on multiple training samples, brain image is obtained and obtains neural network model, can wrap
It includes: original neural network model being trained based on multiple second training samples, obtain preliminary brain image and obtain neural network
Model;Neural network model is obtained to preliminary brain image based on multiple first training samples to be trained, and obtains brain image acquisition
Neural network model.
Wherein, when training sample includes the first training sample and the second training sample, it is assumed that the first training sample is several
Measure the less and preferable sample of training effect, the second training sample is that quantity is more and the general sample of training effect, in order to same
When meet the requirement of training effect and frequency of training, the method that transfer learning can be used.For example, multiple second can be based on
Training sample carries out initial training to original neural network model, obtains preliminary brain image and obtains neural network model;Then, base
Neural network model is obtained to preliminary brain image in multiple first training samples to be trained again, it optionally, can be to preliminary
Brain image obtain neural network model it is rear it is several layers of be trained, can also to preliminary brain image obtain neural network model it is each
Layer is trained.Above-mentioned steps setting is advantageous in that, in the training effect of training sample preferably and in the case where negligible amounts,
Training effect and frequency of training can be met simultaneously, obtain can the covering a variety of situations and preferable brain image of training effect obtains
Take neural network model.
A kind of optional technical solution, sample object may include first sample object;First training sample can pass through
Following steps obtain: obtaining the based on PET-Positron emission computed tomography technology or Single photon emission computer tomography technology
The first sample brain image of one sample object, and, the magnetic resonance sample of first sample object is obtained based on mr imaging technique
This brain image;Magnetic resonance samples brain image and preset standard brain image are registrated, obtain first sample deformation field, and will
First sample brain image and first sample deformation field are as one group of first training sample.
Wherein, include more structural information in magnetic resonance samples brain image, obtained after being registrated with standard brain image
First sample deformation field also include more structural information, when such deformation field is as training sample it is available preferably
Training effect.It therefore, can be using first sample brain image and first sample deformation field as one group of first training sample.It is optional
, when the quantity of magnetic resonance samples brain image is more, multiple first training samples can be directly based upon to original neural network
Model is trained, and is obtained brain image and is obtained neural network model.Above-mentioned steps setting is advantageous in that, can to have trained
Complete brain image, which obtains in neural network model, has structural information, also has structure according to the deformation field that above-mentioned model obtains
Information.
A kind of optional technical solution, sample object may include the second sample object;Second training sample can pass through
Following steps obtain: obtaining the based on PET-Positron emission computed tomography technology or Single photon emission computer tomography technology
Second sample brain image of two sample objects;Second sample brain image and preset standard brain image are registrated, obtain
Two sample deformation fields, and using the second sample brain image and the second sample deformation field as one group of second training sample.
Wherein, first sample object and the second sample object can be identical sample object, be also possible to different samples
This object.It can be obtained after needing subject to receive additional MR scanning due to magnetic resonance samples brain image, then in medical domain
Magnetic resonance samples brain image quantity it is relatively limited.Correspondingly, the second sample brain image only needs subject to receive conventional PET
It can be obtained after scanning or SPECT scanning, then the quantity of the second sample brain image in medical domain is more.It therefore, can be with
Using the second sample brain image and the second sample deformation field as one group of second training sample.Although it is understood that the second instruction
The quantity for practicing sample is more, but does not have structural information in the second sample brain image, so that complete brain image has been trained to obtain
It takes and does not also have structural information in neural network model, training effect is bad.Therefore, the second training sample and the first training sample
It can be used cooperatively, to meet the requirement of training effect and frequency of training simultaneously.
It should be noted that " first sample object " and " the second sample object ", " first sample in the embodiment of the present invention
Brain image " and " the second sample brain image ", " the first training sample " and " the second training sample ", " first sample deformation field " and
" first " and " second " in the similar statement such as " second sample deformation field " is only used for distinguishing each noun, not to each
The restriction of content.
Embodiment three
Fig. 4 is a kind of flow chart of the brain image acquisition methods provided in the embodiment of the present invention three.The present embodiment is with above-mentioned
It is optimized based on each technical solution.In the present embodiment, optionally, using sample brain image and sample deformation field as one
After group training sample, above-mentioned brain image acquisition methods can also include: based on preset data set amplification algorithm to by multiple
The training sample set that training sample is constituted is expanded;Correspondingly, described be based on multiple training samples to original neural network mould
Type is trained, and is obtained brain image and is obtained neural network model, can specifically include: based on the training sample set expanded to original
Beginning neural network model is trained, and is obtained brain image and is obtained neural network model.Wherein, or phase identical as the various embodiments described above
Details are not described herein for the explanation for the term answered.
As shown in figure 4, the method for the present embodiment can specifically include following steps:
S310, the sample brain image for obtaining sample object, sample brain image and preset standard brain image are registrated,
Sample deformation field is obtained, and using sample brain image and sample deformation field as one group of training sample.
S320, the training sample set being made of multiple training samples is expanded based on preset data set amplification algorithm
Increase.
Wherein, it since the training process of original neural network model has higher requirements to the quantity of training sample, that is, trains
The quantity of sample is bigger, and training effect is better.But the particularity of medical domain determines that the quantity of training sample is limited
, i.e. the quantity of sample brain image is limited.Even moreover, identical scanning technique, the sample brain that distinct device generates
The quality of image also has differences, and cannot achieve the purpose of image standardization.Therefore, it is necessary to be calculated based on the amplification of preset data set
Method expands existing training sample set, for example, can be by the methods of translation, overturning, plus noise from having trained sample
The new training sample of a batch is createed in this, obtains neural network model to obtain training effect more preferably brain image.
S330, original neural network model is trained based on the training sample set expanded, obtains brain image acquisition
Neural network model, wherein original neural network model includes convolutional neural networks model.
S340, based on PET-Positron emission computed tomography technology or Single photon emission computer tomography technology obtain by
The first brain image of inspection person.
S350, acquisition have trained complete brain image to obtain neural network model, and the first brain image is input to brain image
Neural network model is obtained, determines the corresponding deformation field of the first brain image.
S360, the first brain image is converted according to deformation field, obtain the second brain image, using the second brain image as by
The brain image of inspection person.
The technical solution of the embodiment of the present invention expands algorithm to the training sample being made of multiple training samples by data set
This collection is expanded, and is trained based on the training sample set expanded to original neural network model, available complete
Property more preferably brain image obtain neural network model.
A kind of optional technical solution, the training based on preset data set amplification algorithm to being made of multiple training samples
Sample set is expanded, and may include: at least two sample deformation fields concentrated according to training sample, and, it is preset random
Deformation field generates amplification deformation field;According to every group of training sample that training sample is concentrated, acquisition has converted sample brain image, and
The generation of sample brain image, which has been converted, according at least two has converted amplification brain image;Become according to having converted amplification brain image and having expanded
Shape field generates non-transformed amplification brain image.
Wherein, firstly, concentrating at least two sample deformation fields of any acquisition from existing training sample, it is based on preset number
It learns average algorithm and solves the mathematic(al) mean of at least two sample deformation field, while preset random deformation field is added, generate
Expand deformation field.Optionally, preset mathematic(al) mean algorithm can be the random weight mathematic(al) mean algorithm of weighted value, count
Average algorithm, geometrical average method etc..Rotational deformation field that preset random deformation field can be randomly generated, translation transformation field
Deng.
Secondly, according to training sample concentrate every group of training sample in sample brain image and sample deformation field, obtain pair
The sample brain image of transformation answered.From existing multiple converted at least two are arbitrarily chosen in sample brain image sample has been converted
Brain image solves described at least two mathematic(al) means for having converted sample brain image based on preset mathematic(al) mean algorithm, generates
Amplification brain image is converted.
Again, according to amplification brain image and amplification deformation field has been converted, non-transformed amplification brain image is generated, and amplification is become
Shape field is added to training sample and concentrates using non-transformed amplification brain image as sample brain image as sample deformation field.
Optionally, the process of above-mentioned training sample set amplification can be expressed by following formula:
Assuming that training sample concentrates shared N group training sample, i-th group of training sample includes sample brain image XiBecome with sample
Shape field Di, i ∈ [1, N].It is corresponding with i-th group of training sample to have converted sample brain image Yi=f (Xi,Di), wherein function f is
Image transform processes.Due to the inverse transformation D of sample deformation fieldi -1It can be according to sample deformation field DiCalculating is got, it may be considered that
Sample brain image Xi=g (Yi,Di -1), wherein function g is image inverse transformation process.Then, according to sample deformation fieldSample becomes
Shape fieldAnd preset random deformation field T, amplification deformation field O can be generated:
Wherein, α, β and γ are the random numbers between [0,1];K1, K2It is the random integers between [1, N].
In turn, according to having converted sample brain imageSample brain image is convertedIt can be generated and converted amplification brain
ImageAnd according to amplification brain image and amplification deformation field O has been converted, non-transformed amplification can be generated
Brain image I:
Wherein, δ is the random number between [0,1];K3, K4It is the random integers between [1, N].
Specifically, K1, K2, K3And K4Can be equal, can part it is equal, can also be not mutually equal.It can by foregoing teachings
Know, only works as K1=K3, when α=δ=1, β=γ=0, the purpose of amplification is not implemented in training sample set;In the case of remaining, training
Sample set realizes the purpose of amplification.Above-mentioned steps setting is advantageous in that, the quantity of training sample can be expanded several times.
Example IV
Fig. 5 is the structural block diagram for the brain image acquisition device that the embodiment of the present invention four provides, and the device is above-mentioned for executing
Brain image acquisition methods provided by any embodiment.The device and the brain image acquisition methods of the various embodiments described above belong to same
A inventive concept, the detail content of not detailed description in the embodiment of brain image acquisition device can refer to above-mentioned brain image
The embodiment of acquisition methods.Referring to Fig. 5, the device is specific can include: the first brain image obtains module 410, model obtains module
420, deformation field determining module 430 and brain image obtain module 440.
Wherein, the first brain image obtains module 410, for being based on PET-Positron emission computed tomography technology or single photon
The first brain image of emission computerized tomography technology acquisition subject;
Model obtains module 420, has trained complete brain image to obtain neural network model for obtaining;
Deformation field determining module 430 obtains neural network model for the first brain image to be input to brain image, determines the
The corresponding deformation field of one brain image;
Brain image obtains module 440, for being converted according to deformation field to the first brain image, the second brain image of acquisition,
Using the second brain image as the brain image of subject.
Optionally, on the basis of above-mentioned apparatus, which may also include that
Training sample obtains module, for obtaining the sample brain image of sample object, by sample brain image and preset mark
Quasi- brain image is registrated, and obtains sample deformation field, and using sample brain image and sample deformation field as one group of training sample;
Model training module obtains mind map for being trained based on multiple training samples to original neural network model
As obtaining neural network model, wherein original neural network model includes convolutional neural networks model.
Optionally, on the basis of above-mentioned apparatus, which further includes model training module, which is used for:
Magnetic resonance samples brain image and PET-Positron emission computed tomography sample brain image based on sample object;And
Preset standard brain image, is trained original neural network model, obtains brain image and obtains neural network model;Or,
Magnetic resonance samples brain image and Single photon emission computer tomography sample brain image based on sample object;And
Preset standard brain image, is trained original neural network model, obtains brain image and obtains neural network model.
Optionally, training sample includes the first training sample and the second training sample;Correspondingly, model training module, tool
Body may include:
Rudimentary model training unit, for being trained based on multiple second training samples to original neural network model,
It obtains preliminary brain image and obtains neural network model;
Model training unit is carried out for obtaining neural network model to preliminary brain image based on multiple first training samples
Training obtains brain image and obtains neural network model.
Optionally, sample object includes first sample object;Training sample obtains module, may include:
The image acquisition unit of first sample object, for being based on PET-Positron emission computed tomography technology or single photon
Emission computerized tomography technology obtains the first sample brain image of first sample object, and, it is based on mr imaging technique
Obtain the magnetic resonance samples brain image of first sample object;
First training sample acquiring unit, for matching magnetic resonance samples brain image and preset standard brain image
Standard obtains first sample deformation field, and using first sample brain image and first sample deformation field as one group of first training sample.
Optionally, sample object includes the second sample object;Training sample obtains module, may include:
The image acquisition unit of second sample object, for being based on PET-Positron emission computed tomography technology or single photon
Emission computerized tomography technology obtains the second sample brain image of the second sample object;
Second training sample acquiring unit, for the second sample brain image and preset standard brain image to be registrated,
The second sample deformation field is obtained, and using the second sample brain image and the second sample deformation field as one group of second training sample.
Optionally, on the basis of above-mentioned apparatus, which further includes sample amplification module, which is used for: based on pre-
If data set amplification algorithm the training sample set being made of multiple training samples is expanded;Correspondingly, model training mould
Block specifically can be used for being trained original neural network model based on the training sample set expanded, obtain brain image and obtain
Take neural network model.
Optionally, sample expands module, may include:
Amplification deformation field generation unit, at least two sample deformation fields for being concentrated according to training sample, and, it presets
Random deformation field, generate amplification deformation field;
Amplification brain image generation unit, every group of training sample for concentrating according to training sample are converted, acquisition has become
This brain image is varied, the generation of sample brain image has been converted according at least two and has converted amplification brain image;
Non-transformed amplification brain image generation unit, for generating not according to amplification brain image and amplification deformation field has been converted
Transformation amplification brain image.
Optionally, on the basis of above-mentioned apparatus, which may also include that
Model modification module, for being adjusted according to the adjustment operation of the user's input received to deformation field, and base
Neural network model is obtained to brain image in deformation field adjusted and the first brain image to be trained again, and is tied according to training
Fruit updates brain image and obtains neural network model.
Optionally, brain image obtains module 440, specifically can be used for: according to deformation field to the target in the first brain image
The position coordinates of pixel are converted, and the second brain image is obtained.
The brain image acquisition device that the embodiment of the present invention four provides obtains module by the first brain image, model obtains mould
First brain image is input to brain image and obtained in neural network model, determined by the mutual cooperation of block and deformation field determining module
The corresponding deformation field of first brain image;Obtaining module by brain image converts deformation field to the first brain image, obtains
Second brain image, using the second brain image as the brain image of subject.Above-mentioned apparatus solves process of image registration bring essence
The problem of degree limitation, improve the structure precision of the brain image of subject;Moreover, subject is not necessarily to carry out additional MR scanning,
It is more convenient to operate.
Mind map provided by any embodiment of the invention can be performed in brain image acquisition device provided by the embodiment of the present invention
As acquisition methods, have the corresponding functional module of execution method and beneficial effect.
It is worth noting that, in the embodiment of above-mentioned brain image acquisition device, included each unit and module are
It is divided according to the functional logic, but is not limited to the above division, as long as corresponding functions can be realized;Separately
Outside, the specific name of each functional unit is also only for convenience of distinguishing each other, the protection scope being not intended to restrict the invention.
Embodiment five
Fig. 6 is a kind of structural schematic diagram for equipment that the embodiment of the present invention five provides, as shown in fig. 6, the equipment includes depositing
Reservoir 510, processor 520, input unit 530 and output device 540.The quantity of processor 520 in equipment can be one
Or it is multiple, in Fig. 6 by taking a processor 520 as an example;Memory 510, processor 520, input unit 530 and output in equipment
Device 540 can be connected by bus or other means, in Fig. 6 for being connected by bus 550.
Memory 510 is used as a kind of computer readable storage medium, can be used for storing software program, journey can be performed in computer
Sequence and module, if the corresponding program instruction/module of brain image acquisition methods in the embodiment of the present invention is (for example, brain image obtains
It takes the first brain image in device to obtain module 410, model acquisition module 420, deformation field determining module 430 and brain image to obtain
Module 440).Software program, instruction and the module that processor 520 is stored in memory 510 by operation, thereby executing setting
Standby various function application and data processing realizes above-mentioned brain image acquisition methods.
Memory 510 can mainly include storing program area and storage data area, wherein storing program area can store operation system
Application program needed for system, at least one function;Storage data area, which can be stored, uses created data etc. according to equipment.This
Outside, memory 510 may include high-speed random access memory, can also include nonvolatile memory, for example, at least one
Disk memory, flush memory device or other non-volatile solid state memory parts.In some instances, memory 510 can be into one
Step includes the memory remotely located relative to processor 520, these remote memories can pass through network connection to equipment.On
The example for stating network includes but is not limited to internet, intranet, local area network, mobile radio communication and combinations thereof.
Input unit 530 can be used for receiving the number or character information of input, and generate with the user setting of device with
And the related key signals input of function control.Output device 540 may include that display screen etc. shows equipment.
Embodiment six
The embodiment of the present invention six provides a kind of storage medium comprising computer executable instructions, and the computer is executable
When being executed by computer processor for executing a kind of brain image acquisition methods, this method may include: based on positive electricity for instruction
Sub- emission computerized tomography technology or Single photon emission computer tomography technology obtain the first brain image of subject;It obtains
Complete brain image has been trained to obtain neural network model;First brain image is input to brain image and obtains neural network model,
Determine the corresponding deformation field of the first brain image;The first brain image is converted according to deformation field, obtains the second brain image, by the
Brain image of two brain images as subject.
Certainly, a kind of storage medium comprising computer executable instructions, computer provided by the embodiment of the present invention
The method operation that executable instruction is not limited to the described above, can also be performed brain image provided by any embodiment of the invention and obtains
Take the relevant operation in method.
By the description above with respect to embodiment, it is apparent to those skilled in the art that, the present invention
It can be realized by software and required common hardware, naturally it is also possible to which by hardware realization, but in many cases, the former is more
Good embodiment.According to such understanding, what technical solution of the present invention substantially in other words contributed to the prior art
Part can be embodied in the form of software products, which can store in computer readable storage medium
In, floppy disk, read-only memory (Read-Only Memory, ROM), random access memory (Random such as computer
Access Memory, RAM), flash memory (FLASH), hard disk or CD etc., including some instructions are with so that a computer is set
Standby (can be personal computer, server or the network equipment etc.) executes method described in each embodiment of the present invention.
Note that the above is only a better embodiment of the present invention and the applied technical principle.It will be appreciated by those skilled in the art that
The invention is not limited to the specific embodiments described herein, be able to carry out for a person skilled in the art it is various it is apparent variation,
It readjusts and substitutes without departing from protection scope of the present invention.Therefore, although being carried out by above embodiments to the present invention
It is described in further detail, but the present invention is not limited to the above embodiments only, without departing from the inventive concept, also
It may include more other equivalent embodiments, and the scope of the invention is determined by the scope of the appended claims.
Claims (13)
1. a kind of brain image acquisition methods characterized by comprising
The of subject is obtained based on PET-Positron emission computed tomography technology or Single photon emission computer tomography technology
One brain image;
It obtains and complete brain image has been trained to obtain neural network model;
First brain image is input to the brain image and obtains neural network model, determines that first brain image is corresponding
Deformation field;
First brain image is converted according to the deformation field, obtains the second brain image, second brain image is made
For the brain image of subject.
2. the method according to claim 1, wherein the method also includes:
The sample brain image for obtaining sample object, the sample brain image and preset standard brain image are registrated, obtained
Sample deformation field, and using the sample brain image and the sample deformation field as one group of training sample;
Original neural network model is trained based on multiple training samples, the brain image is obtained and obtains neural network
Model, wherein the original neural network model includes convolutional neural networks model.
3. the method according to claim 1, wherein the method also includes:
Magnetic resonance samples brain image and PET-Positron emission computed tomography sample brain image based on sample object;And it is default
Standard brain image, original neural network model is trained, the brain image is obtained and obtains neural network model;Or,
Magnetic resonance samples brain image and Single photon emission computer tomography sample brain image based on the sample object;And
Preset standard brain image, is trained original neural network model, obtains the brain image and obtains neural network model.
4. according to the method described in claim 2, it is characterized in that, the training sample includes the first training sample and the second instruction
Practice sample;Correspondingly, described be trained original neural network model based on multiple training samples, the mind map is obtained
As obtaining neural network model, comprising:
Original neural network model is trained based on multiple second training samples, preliminary brain image is obtained and obtains nerve
Network model;
Neural network model is obtained to the preliminary brain image based on multiple first training samples to be trained, and is obtained described
Brain image obtains neural network model.
5. according to the method described in claim 4, it is characterized in that, the sample object includes first sample object;Described
One training sample obtains as follows:
First sample is obtained based on PET-Positron emission computed tomography technology or Single photon emission computer tomography technology
The first sample brain image of this object, and, the magnetic resonance sample of the first sample object is obtained based on mr imaging technique
This brain image;
The magnetic resonance samples brain image and preset standard brain image are registrated, obtain first sample deformation field, and will
The first sample brain image and the first sample deformation field are as one group of first training sample.
6. method according to claim 4 or 5, which is characterized in that the sample object includes the second sample object;It is described
Second training sample obtains as follows:
Second sample is obtained based on PET-Positron emission computed tomography technology or Single photon emission computer tomography technology
The second sample brain image of this object;
The second sample brain image and preset standard brain image are registrated, obtain the second sample deformation field, and by institute
The second sample brain image and second sample deformation field are stated as one group of second training sample.
7. according to the method described in claim 2, it is characterized in that, described by the sample brain image and the sample deformation
After field is as one group of training sample, the method also includes: based on preset data set amplification algorithm to by multiple instructions
Practice the training sample set that sample is constituted to be expanded;
Correspondingly, described be trained original neural network model based on multiple training samples, the brain image is obtained
Obtain neural network model, comprising:
Original neural network model is trained based on the training sample set expanded, the brain image is obtained and obtains nerve net
Network model.
8. the method according to the description of claim 7 is characterized in that described expand algorithm to by multiple based on preset data set
The training sample set that the training sample is constituted is expanded, comprising:
According at least two sample deformation fields that the training sample is concentrated, and, preset random deformation field generates and expands
Increase deformation field;
According to every group of training sample that the training sample is concentrated, acquisition has converted sample brain image, and according at least two institutes
State converted sample brain image generation converted amplification brain image;
Amplification brain image and the amplification deformation field have been converted according to described, has generated non-transformed amplification brain image.
9. the method according to claim 1, wherein the method also includes:
The deformation field is adjusted according to the adjustment operation of the user's input received, and based on deformation field adjusted and
First brain image obtains neural network model to the brain image and is trained again, and described in being updated according to training result
Brain image obtains neural network model.
10. the method according to claim 1, wherein it is described according to the deformation field to first brain image
It is converted, obtains the second brain image, comprising:
It is converted according to position coordinates of the deformation field to the target pixel points in first brain image, obtains the second brain
Image.
11. a kind of acquisition device of brain image characterized by comprising
First brain image obtains module, for being based on PET-Positron emission computed tomography technology or single photon emission computer
The first brain image of tomographic techniques acquisition subject;
Model obtains module, has trained complete brain image to obtain neural network model for obtaining;
Deformation field determining module obtains neural network model for first brain image to be input to the brain image, determines
The corresponding deformation field of first brain image;
Brain image obtains module, for being converted according to the deformation field to first brain image, the second brain image of acquisition,
Using second brain image as the brain image of subject.
12. a kind of equipment, which is characterized in that the equipment includes:
One or more processors;
Memory, for storing one or more programs;
When one or more of programs are executed by one or more of processors, so that one or more of processors are real
Now following operation:
The of subject is obtained based on PET-Positron emission computed tomography technology or Single photon emission computer tomography technology
One brain image;
It obtains and complete brain image has been trained to obtain neural network model;
First brain image is input to the brain image and obtains neural network model, determines that first brain image is corresponding
Deformation field;
First brain image is converted according to the deformation field, obtains the second brain image, second brain image is made
For the brain image of subject.
13. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that the computer program
Following operation is realized when being executed by processor:
The of subject is obtained based on PET-Positron emission computed tomography technology or Single photon emission computer tomography technology
One brain image;
It obtains and complete brain image has been trained to obtain neural network model;
First brain image is input to the brain image and obtains neural network model, determines that first brain image is corresponding
Deformation field;
First brain image is converted according to the deformation field, obtains the second brain image, second brain image is made
For the brain image of subject.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910260795.2A CN109961435B (en) | 2019-04-02 | 2019-04-02 | Brain image acquisition method, device, equipment and storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910260795.2A CN109961435B (en) | 2019-04-02 | 2019-04-02 | Brain image acquisition method, device, equipment and storage medium |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109961435A true CN109961435A (en) | 2019-07-02 |
CN109961435B CN109961435B (en) | 2022-02-22 |
Family
ID=67025519
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910260795.2A Active CN109961435B (en) | 2019-04-02 | 2019-04-02 | Brain image acquisition method, device, equipment and storage medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109961435B (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113192014A (en) * | 2021-04-16 | 2021-07-30 | 深圳市第二人民医院(深圳市转化医学研究院) | Training method, device, electronic equipment and medium for improving ventricle segmentation model |
CN113450397A (en) * | 2021-06-25 | 2021-09-28 | 广州柏视医疗科技有限公司 | Image deformation registration method based on deep learning |
WO2021139835A3 (en) * | 2021-03-24 | 2021-11-11 | 苏州深透智能科技有限公司 | Spect imaging prediction model creation method and apparatus, and device and storage medium |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7945117B2 (en) * | 2006-08-22 | 2011-05-17 | Siemens Medical Solutions Usa, Inc. | Methods and systems for registration of images |
EP3121789A1 (en) * | 2015-07-22 | 2017-01-25 | Siemens Medical Solutions USA, Inc. | Method and system for convolutional neural network regression based 2d/3d image registration |
CN106920228A (en) * | 2017-01-19 | 2017-07-04 | 北京理工大学 | The method for registering and device of brain map and brain image |
CN109102532A (en) * | 2017-06-20 | 2018-12-28 | 西门子保健有限责任公司 | The metaplasia of deep learning for medical imaging |
US20190050999A1 (en) * | 2017-08-14 | 2019-02-14 | Siemens Healthcare Gmbh | Dilated Fully Convolutional Network for Multi-Agent 2D/3D Medical Image Registration |
CN109427058A (en) * | 2017-08-17 | 2019-03-05 | 西门子保健有限责任公司 | Automatic variation detection in medical image |
-
2019
- 2019-04-02 CN CN201910260795.2A patent/CN109961435B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7945117B2 (en) * | 2006-08-22 | 2011-05-17 | Siemens Medical Solutions Usa, Inc. | Methods and systems for registration of images |
EP3121789A1 (en) * | 2015-07-22 | 2017-01-25 | Siemens Medical Solutions USA, Inc. | Method and system for convolutional neural network regression based 2d/3d image registration |
CN106920228A (en) * | 2017-01-19 | 2017-07-04 | 北京理工大学 | The method for registering and device of brain map and brain image |
CN109102532A (en) * | 2017-06-20 | 2018-12-28 | 西门子保健有限责任公司 | The metaplasia of deep learning for medical imaging |
US20190050999A1 (en) * | 2017-08-14 | 2019-02-14 | Siemens Healthcare Gmbh | Dilated Fully Convolutional Network for Multi-Agent 2D/3D Medical Image Registration |
CN109427058A (en) * | 2017-08-17 | 2019-03-05 | 西门子保健有限责任公司 | Automatic variation detection in medical image |
Non-Patent Citations (4)
Title |
---|
ARISTEIDIS SOTIRAS ET AL: "Deformable Medical Image Registration: A Survey", 《IEEE TRANSACTIONS ON MEDICAL IMAGING》 * |
HRISTINA UZUNOVA ET AL: "Training CNNs for image registration from few samples with model-based data augmentation", 《MEDICAL IMAGE COMPUTING AND COMPUTER ASSISTED INTERVENTION − MICCAI 2017》 * |
宋斐: "医学图像配准框架的研究进展及展望", 《电子技术与软件工程》 * |
陆雪松等: "一种面向医学图像非刚性配准的多维特征度量方法", 《自动化学报》 * |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2021139835A3 (en) * | 2021-03-24 | 2021-11-11 | 苏州深透智能科技有限公司 | Spect imaging prediction model creation method and apparatus, and device and storage medium |
CN113192014A (en) * | 2021-04-16 | 2021-07-30 | 深圳市第二人民医院(深圳市转化医学研究院) | Training method, device, electronic equipment and medium for improving ventricle segmentation model |
CN113192014B (en) * | 2021-04-16 | 2024-01-30 | 深圳市第二人民医院(深圳市转化医学研究院) | Training method and device for improving ventricle segmentation model, electronic equipment and medium |
CN113450397A (en) * | 2021-06-25 | 2021-09-28 | 广州柏视医疗科技有限公司 | Image deformation registration method based on deep learning |
CN113450397B (en) * | 2021-06-25 | 2022-04-01 | 广州柏视医疗科技有限公司 | Image deformation registration method based on deep learning |
Also Published As
Publication number | Publication date |
---|---|
CN109961435B (en) | 2022-02-22 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN112017198B (en) | Right ventricle segmentation method and device based on self-attention mechanism multi-scale features | |
EP4148660B1 (en) | Improving quality of medical images using multi-contrast and deep learning | |
Dangi et al. | A distance map regularized CNN for cardiac cine MR image segmentation | |
CN113099208A (en) | Method and device for generating dynamic human body free viewpoint video based on nerve radiation field | |
CN107909622B (en) | Model generation method, medical imaging scanning planning method and medical imaging system | |
CN110444277B (en) | Multi-mode brain MRI image bidirectional conversion method based on multi-generation and multi-confrontation | |
CN107133996A (en) | Produce the method and PET/CT systems for the decay pattern rebuild for PET data | |
CN109961435A (en) | Brain image acquisition methods, device, equipment and storage medium | |
US8861891B2 (en) | Hierarchical atlas-based segmentation | |
WO2024021523A1 (en) | Graph network-based method and system for fully automatic segmentation of cerebral cortex surface | |
CN116402865B (en) | Multi-mode image registration method, device and medium using diffusion model | |
EP3555850A1 (en) | System and method for image segmentation using a joint deep learning model | |
CN111860782B (en) | Triple multi-scale CycleGAN, fundus fluorography generation method, computer device, and storage medium | |
CN112819914A (en) | PET image processing method | |
CN111814891A (en) | Medical image synthesis method, device and storage medium | |
WO2021184195A1 (en) | Medical image reconstruction method, and medical image reconstruction network training method and apparatus | |
CN110580509B (en) | Multimodal data processing system and method for generating countermeasure model based on hidden representation and depth | |
Tilborghs et al. | Shape constrained cnn for cardiac mr segmentation with simultaneous prediction of shape and pose parameters | |
CN114241240A (en) | Method and device for classifying brain images, electronic equipment and storage medium | |
CN113870327A (en) | Medical image registration method based on multi-level deformation field prediction | |
CN116152235A (en) | Cross-modal synthesis method for medical image from CT (computed tomography) to PET (positron emission tomography) of lung cancer | |
WO2022120731A1 (en) | Mri-pet image modality conversion method and system based on cyclic generative adversarial network | |
CN115861464A (en) | Pseudo CT (computed tomography) synthesis method based on multimode MRI (magnetic resonance imaging) synchronous generation | |
CN116168097A (en) | Method, device, equipment and medium for constructing CBCT sketching model and sketching CBCT image | |
CN111862320B (en) | Automatic steering method for SPECT three-dimensional reconstruction image to standard view |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
CB02 | Change of applicant information | ||
CB02 | Change of applicant information |
Address after: 201807 Shanghai City, north of the city of Jiading District Road No. 2258 Applicant after: Shanghai Lianying Medical Technology Co., Ltd Address before: 201807 Shanghai City, north of the city of Jiading District Road No. 2258 Applicant before: SHANGHAI UNITED IMAGING HEALTHCARE Co.,Ltd. |
|
GR01 | Patent grant | ||
GR01 | Patent grant |